AI on IBM Z & IBM LinuxONE - Group home

TensorFlow v2.3 Image Frequently Asked Questions (FAQ)

  

By Gabriel Borger, IBM Z Data & AI - Offering Manager


FAQ for the recently announce  New TensorFlow v2.3 Container Image for Linux on IBM Z and LinuxONE!

Where can I get the TensorFlow image?

TensorFlow v2.3 for IBM Z/LinuxONE can be found on the IBM Z Container Registry. You can go to the Container Registry index page of the IBM Z and LinuxONE Container Image Registry and navigate to TensorFlow v2.3 to install the image. Be sure you have the IBM Z/LinuxONE TensorFlow utility running alongside your TensorFlow applications and environments if you plan to run models built on little endian (i.e. x86) platforms.

 

How do I install TensorFlow on my IBM Z or LinuxONE machine?

You can simply use the TensorFlow Linux on Z docker image found on the IBM Cloud Repository. Docker on Linux on Z is just like docker on other platforms. General guidance on docker on Linux on Z is available here. TensorFlow may also be built from source

 

Can I train models on IBM Z?

Yes. All you need is the TensorFlow base image installed in a linux environment (either zCX or Linux on Z) and access to the data you wish to train your models with. You can learn more about building a training model from this tutorial.

You can also train models on your local device and deploy them on your Linux on IBM Z, LinuxONE, or zCX environments. Check out this tutorial on how to train your models and deploy with TensorFlow serving. TensorFlow Serving build instructions can be found here. Be sure to execute the IBM Z/LinuxONE TensorFlow utility installed alongside your TensorFlow applications and environments if you plan to run models built on little endian (i.e. x86) platforms.

 

How do I port a model?

As with other platforms, portability relies on the save and load model APIs provided by TensorFlow. 

Guidance can be found here. When deploying a model that was trained on an little endian environment (such as x86) to IBM Z, you will also execute the IBM Z/LinuxONE TensorFlow utility to ensure the model can execute on Z.

 

Does it run on all Linux flavors?

TensorFlow docker or podman images for s390x will largely be portable to most Linux on IBM Z distributions.  Linux on IBM Z Ecosystem team maintains a validated TensorFlow port for s390x along with build guidance which can be found here.

 

Can it run on z/OS native?

No, TensorFlow has not yet been ported to z/OS natively, however it can be deployed to z/OS container extensions (zCX). zCX allows a client to deploy a TensorFlow docker container collocated with z/OS workloads.  

 

Can it run on zCX?

TensorFlow and TensorFlow ecosystem docker containers can be deployed to z/OS Container Extensions. zCX allows a client to deploy a TensorFlow docker container collocated with z/OS workloads. 

 

Is any of the TensorFlow ecosystem supported on Linux on Z?

TensorFlow Serving and TensorFlow Transform are ported and validated to run on Linux on Z/LinuxONE. Other ecosystem packages may be supported. 

 

Is there an official support? How can I provide comments or report issues?

There is currently no official support option for TensorFlow on IBM Z/LinuxONE, however an official support option is currently in development. To stay up to date on service and support options, please fill out this form. Community support is available through the TensorFlow on IBM Z/LinuxONE Github community.

 

How can I become a sponsor user? Are there any benefits to being a sponsor user?

Our sponsor users help our team better understand how our clients use our products. As a sponsor user, our offering, design, and development teams will work with you to better understand how to improve our products and experiences to ultimately better serve you and the rest of our users.  If you are interested in joining our sponsor user program, please fill out this form or contact either Karina Campos (karina.campos@ibm.com) or Christopher Buchholz (buchholz@us.ibm.com) for more information.

 

Can IBM help me understand how TensorFlow fits into my business? 

The IBM Garage for Systems offers demos and working sessions to help clients understand how to leverage TensorFlow's machine learning (ML) and deep learning (DL) capabilities to augment, automate, or expand upon key workloads. You can learn more about the IBM Garage for Systems here. If you are interested in working with our garage, feel free to contact SysGarage@us.ibm.com.

 

Where can I learn more about machine learning (ML), deep learning (DL) and TensorFlow?

IBM offers various resources for those interested in learning about ML and DL including various blogs and videos linked below. 

 

 

How well does TensorFlow perform on IBM Z and LinuxONE? Do I need any specific resources or skills?

IBM Z and LinuxONE servers allow for excellent scalability of all workloads, including TensorFlow. For direct TensorFlow users, no specific additional skills are required.

 

Who can I reach out to if I have more questions?

If you have any questions about the TensorFlow v2.3 image for IBM Z/LinuxONE, feel free to reach out to aionz@us.ibm.